Zurich AI Festival: Google Shows What AI Can Do

As part of the first Zurich AI Festival, Google offered insights into cutting-edge research and applications of artificial intelligence at its Hürlimann Campus. Engineers presented state-of-the-art projects in a series of lightning talks—ranging from cybersecurity and image recognition to climate research and accessibility. The event served as a platform for exchange between local talent and global innovation, underscoring Zurich’s growing importance as an AI hub.

Google in Zurich: A Driving Force Behind the Ecosystem

Christine Antlanger-Winter, Country Director for Google in Switzerland, opened the event and emphasized Google’s deep roots in Zurich. The company has been present here for over 20 years and now employs 5,000 people—affectionately known as „Zooglers“—across two campuses. She emphasized that Google sees itself as an active part of Zurich’s thriving tech ecosystem.

This is evident in long-standing partnerships, such as the one with the Greater Zurich Area, whose office is now located in the very spot where the first two Google employees got started in 2004. The company is particularly proud of the fact that former Google employees have already founded 110 startups, creating 1,700 jobs. In addition, Switzerland was the first country where Google launched an apprenticeship program, which now serves as a model for other locations.

Sabine Müller, deputy managing director of the Greater Zurich Area and co-initiator of the Zurich AI Festival, explained the festival’s origins and goals. The festival aims to position Zurich and Switzerland as a global beacon for ethical, impactful, and inclusive AI. The mission, she said, is to make AI accessible and to foster public dialogue about its role in the future. Müller described how Zurich has evolved over the past 20 years from a city once derided as „boring“ into a global innovation hub with a higher concentration of big-tech companies than Silicon Valley.

The idea for the festival arose from a desire to highlight this strength and create a platform for knowledge exchange and collaboration, similar to what other cities are already doing. With 26 partners on board and an expected 5,000 guests, the festival celebrates the concentrated expertise of the local ecosystem. In his opening remarks, Alexander Ilic, Director of the ETH AI Center, emphasized the importance of collaboration between academia and industry and proudly mentioned the release of the first truly open Swiss LLM with 70 billion parameters.

AI in the Shadows: Protection Against Cyberattacks

Daniel Fabian, AI Red Team Manager for Google’s Cybersecurity Team, shed light on the dark side of AI. He explained that new technologies are inevitably exploited by attackers for malicious purposes. Hackers are already using AI as a productivity tool—for example, to create personalized phishing emails more quickly or to autonomously find vulnerabilities in systems. Fabian’s team acts as a „crash test“ unit for AI: They take on the role of attackers to target their own systems and uncover vulnerabilities before others do. The insights gained from these simulated attacks are directly incorporated into the improvement of Google products, such as the Gemini models, and help close entire classes of security vulnerabilities.

Google Lens: Understanding the World Through the Camera

Vanessa Ackermann, Software Engineer at Google, gave an insight into Google Lens, a technology that was invented right here in Zurich.  Google Lens turns your smartphone camera into a versatile search tool that processes over 20 billion queries every month. The fundamental problem that Lens solves is the fact that not everything you see can be easily put into words. Whether it’s an unknown plant, a historic building, or a piece of clothing you saw online—Lens bridges the gap between seeing and searching.

The use cases are extremely diverse:

- Identification: You can point to works of art, landmarks, or products to get more information.

- Translation: Lens can translate text in over 100 languages in real time on the camera screen, which is especially helpful when traveling—for example, with menus or signs.

- Homework Help: Instead of simply giving away the answer, Lens breaks down math and chemistry problems into individual steps to support the learning process. To do this, Google works with a team of educators.

- Text and Shopping Features: You can copy text directly from the real world or take photos of products you see to find purchase options, similar items, and reviews.

- Health Information: Lens can also help identify rashes or moles by displaying visually similar images from the internet to provide initial guidance. However, Google emphasizes that this is not a substitute for medical advice.

The core technology behind Lens is what is known as «embedding», which can be thought of as a digital, numerical fingerprint for an image. An AI model is trained using billions of images from the internet to learn how to generate an optimal fingerprint for each image. These fingerprints are designed so that visually similar images are located close to one another in a multidimensional mathematical space. When a user takes a photo, the model calculates the fingerprint for that image and searches for its nearest neighbors in this space. Based on the descriptions of these neighboring images, Lens can then precisely determine what is shown in the photo, even if it has never seen that exact image before. This method makes it possible to recognize even subtle differences, such as those between very similar plant species.

Project Contrails: AI Against Climate Change

Max Vogler, Software Engineering at Google, presented how Google is using AI to combat an often-overlooked climate factor: contrails. These artificial clouds, which form behind airplanes, are responsible for about one-third of the global warming caused by aviation. An AI model, trained using satellite images, can predict the formation of these warming contrails.

Since contrails form only within very narrow altitude ranges, pilots can prevent them from forming entirely by making slight adjustments to their altitude of just a few thousand feet. A study conducted with American Airlines showed that 54 % of contrails could be avoided with an additional fuel consumption of only 0.3 %. This project is part of Google's goal to offset one gigatonne of CO₂ equivalent annually worldwide starting in 2030.

VEO: The Next Level of AI-Powered Video Generation

Pieter-Jan Kindermans, Staff Research Scientist at Google DeepMind, presented VEO, a groundbreaking AI model that converts text prompts into videos. They demonstrated the impressive progress made since the first experiments in 2022—which produced low-resolution, slightly distorted videos of a teddy bear underwater—up to the current VEO 3 model, which represents a major qualitative breakthrough.

What makes VEO 3 special is its ability to produce not only visually appealing, but also to create narratively coherent and physically plausible scenes—including sound, dialogue, and realistic interactions. As a striking example, a video of a male ibex and his son in the Swiss Alps was shown. The model created the scene based on a story about mountain biker Nino Schurter. VEO 3 autonomously generated details—without explicit instructions—that bring the scene to life:

- Realistic sound: The model generated a lower-pitched voice for the father and a higher-pitched one for the son, added animal sounds between sentences, and incorporated ambient sounds such as footsteps.

- Understanding of Physics: It took into account physical realities such as light reflections in a puddle and the interaction of the hooves with the water.

- Cinematic storytelling: Through the use of camera pans and scene changes—such as zooming in on a tire track—VEO creates a narrative structure reminiscent of television productions.

Pieter-Jan Kindermans emphasized that VEO is designed to give everyone the opportunity to bring their imagination to life and tell captivating stories. He encouraged the audience to check out online the amazing creations users have already made with the tool, since even the developers themselves are constantly surprised by what’s possible with their models.

AI Co-Scientist: AI as a Research Partner in Medicine

Petar Sirkovic, Staff Software Engineer, presented the "AI Co-Scientist" project, which aims to radically accelerate the scientific research process. The Gemini-based multi-agent system is designed to help researchers manage the vast amount of existing knowledge and identify the most promising hypotheses. Various AI agents simulate the scientific process: they generate, criticize, evaluate and improve hypotheses in an evolutionary cycle. In a practical test on antimicrobial resistance, a problem that a human research team had been working on for years, the AI Co-Scientist was given the same research question.

After just two days, the top hypothesis generated by the system was identical to the biological solution that the human team had independently validated in the lab. The lead researcher, Prof. José Penadés, was so stunned that his first reaction was: „You’ve hacked my emails.“.

Ask Studio: The AI Partner for YouTube Creators

Patricia Magro, Product Manager at YouTube, introduced „Ask Studio,“ an AI assistant that is integrated directly into YouTube Studio and is being co-developed by a team in Zurich. Since life as a creator is very demanding and many of them work without a team, Ask Studio is intended to serve as a creative partner who takes on complex tasks and provides sound advice.

What sets Ask Studio apart from an ordinary chatbot is its direct access to a channel's specific data and content. This enables the tool to perform in-depth analyses:

- Complex Data Queries: A creator can ask whether their viewers watch for longer on TV or on mobile devices. Ask Studio then analyzes the relevant channel data to provide an answer.

- Content Analysis: If viewers stop watching at a certain point in a video, Ask Studio can analyze the content and transcript at that point and suggest possible reasons, such as an intro that's too long or an inside joke that new viewers don't understand.

- Semantic Search in Comments: Instead of just searching for keywords, Ask Studio can analyze comments based on their meaning to identify, for example, actionable feedback or ideas for new videos.

- Creative Support: The tool can also suggest titles and outlines for new videos to help streamline the creative process.

Another key goal is to provide creators with reliable information. Ask Studio bases its answers on official YouTube guidelines and help articles to debunk myths about „algorithm hacks“ and shift the focus toward creating great content. With just one click, a creator can get a complete analysis of their latest video’s performance—a process that used to take 20 minutes but now takes just seconds.

AI for Accessibility: Technology That Helps Everyone

Dirk Ginader, Accessibility Engineering Lead at Google, closed the event with a compelling presentation on the role of AI in improving accessibility. He emphasized that over a billion people worldwide have some form of disability. Technologies such as Google Lookout, which are based on the same principles as Lens, enable blind people to „see“ their surroundings and live more independently.

Today, AI can automatically generate detailed image descriptions for people with visual impairments where none previously existed. He also introduced „Project Relate,“ an app that is being trained to understand the unique speech patterns of people with speech impairments and to transcribe their words in real time or repeat them in a clear, computerized voice. He pointed out that many technologies we use every day, such as email or captions, were originally developed for people with disabilities and ultimately benefit everyone.

Alexander Ilic, Director of the ETH AI Center, concluded by emphasizing the need for collaboration to pool and strengthen existing resources. With over 1,400 doctoral students and 120 professors, the ETH AI Center is the largest university-based AI center in Europe and is driving basic research forward, for example with the release of the first truly open Swiss LLM with 70 billion parameters. Ilic highlighted how crucial partnerships with companies like Google are for taking this basic research to the next level and creating innovations that we use every day. The synergy between excellent research, a startup-friendly environment, and established tech companies reinforces their shared success.

The Zurich AI Festival is thus more than just a series of events; it is a clear commitment to the mission of making AI accessible, accelerating collaboration between research and industry, and establishing Zurich as a global beacon for ethical and impactful AI. It is an invitation to network and shape the future together.